git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH 4/6] fetch-pack: optimize loading of refs via commit graph

From
Derrick Stolee <stolee@gmail.com>
Date
Aug 20, 2021, 14:37 UTC
Message-ID
<d280d167-3e62-f73b-5504-9d1d8bdacaab@gmail.com>
In-Reply-To
<67917af7ceeefe41ae0f6edf69cd61e2ee8c0ea3.1629452412.git.ps@pks.im>
On 8/20/2021 6:08 AM, Patrick Steinhardt wrote:
Show 21 quoted lines
> In order to negotiate a packfile, we need to dereference refs to see
> which commits we have in common with the remote. To do so, we first look
> up the object's type -- if it's a tag, we peel until we hit a non-tag
> object. If we hit a commit eventually, then we return that commit.
> 
> In case the object ID points to a commit directly, we can avoid the
> initial lookup of the object type by opportunistically looking up the
> commit via the commit-graph, if available, which gives us a slight speed
> bump of about 2% in a huge repository with about 2.3M refs:
> 
>     Benchmark #1: HEAD~: git-fetch
>       Time (mean ± σ):     31.634 s ±  0.258 s    [User: 28.400 s, System: 5.090 s]
>       Range (min … max):   31.280 s … 31.896 s    5 runs
> 
>     Benchmark #2: HEAD: git-fetch
>       Time (mean ± σ):     31.129 s ±  0.543 s    [User: 27.976 s, System: 5.056 s]
>       Range (min … max):   30.172 s … 31.479 s    5 runs
> 
>     Summary
>       'HEAD: git-fetch' ran
>         1.02 ± 0.02 times faster than 'HEAD~: git-fetch'

This 2% gain is nice, especially because you are measuring the end-to-end scenario. If you use GIT_TRACE2_PERF=1 on a few runs, then you could likely isolate some of the regions from mark_complete_and_common_ref() and demonstrate a larger improvement in that focused area.

Show 9 quoted lines
> @@ -119,6 +119,11 @@ static struct commit *deref_without_lazy_fetch(const struct object_id *oid,
>  {
>  	enum object_type type;
>  	struct object_info info = { .typep = &type };
> +	struct commit *commit;
> +
> +	commit = lookup_commit_in_graph(the_repository, oid);
> +	if (commit)
> +		return commit;
Obviously a correct thing to do.
>  	if (type == OBJ_COMMIT) {
> -		struct commit *commit = lookup_commit(the_repository, oid);
> +		commit = lookup_commit(the_repository, oid);
Re-using the local simplifies this. Good.

Thanks, -Stolee

Previous: Patrick SteinhardtNext: Patrick Steinhardt
Message 15 of 48 in “Speed up mirror-fetches with many refs”
  1. 0/6 Speed up mirror-fetches with many refsPatrick Steinhardt, Aug 20, 2021
  2. 1/6 fetch: speed up lookup of want refs via commit-graphPatrick Steinhardt, Aug 20, 2021
  3. Derrick StoleeAug 20, 2021
  4. Junio C HamanoAug 20, 2021
  5. Patrick SteinhardtAug 23, 2021
  6. Derrick StoleeAug 25, 2021
  7. 2/6 fetch: avoid unpacking headers in object existence checkPatrick Steinhardt, Aug 20, 2021
  8. Ævar Arnfjörð BjarmasonAug 25, 2021
  9. 3/6 connected: refactor iterator to return next object ID directlyPatrick Steinhardt, Aug 20, 2021
  10. Derrick StoleeAug 20, 2021
  11. Junio C HamanoAug 20, 2021
  12. René ScharfeAug 20, 2021
  13. Patrick SteinhardtAug 23, 2021
  14. 4/6 fetch-pack: optimize loading of refs via commit graphPatrick Steinhardt, Aug 20, 2021
  15. Derrick StoleeAug 20, 2021
  16. 5/6 fetch: refactor fetch refs to be more extendablePatrick Steinhardt, Aug 20, 2021
  17. Derrick StoleeAug 20, 2021
  18. 6/6 fetch: avoid second connectivity check if we already have all objectsPatrick Steinhardt, Aug 20, 2021
  19. Derrick StoleeAug 20, 2021
  20. Patrick SteinhardtAug 23, 2021
  21. Derrick StoleeAug 20, 2021
  22. Junio C HamanoAug 21, 2021
  23. 0/7 Speed up mirror-fetches with many refsPatrick Steinhardt, Aug 24, 2021
  24. 1/7 fetch: speed up lookup of want refs via commit-graphPatrick Steinhardt, Aug 24, 2021
  25. Derrick StoleeAug 25, 2021
  26. 2/7 fetch: avoid unpacking headers in object existence checkPatrick Steinhardt, Aug 24, 2021
  27. 3/7 connected: refactor iterator to return next object ID directlyPatrick Steinhardt, Aug 24, 2021
  28. 4/7 fetch-pack: optimize loading of refs via commit graphPatrick Steinhardt, Aug 24, 2021
  29. 5/7 fetch: refactor fetch refs to be more extendablePatrick Steinhardt, Aug 24, 2021
  30. Derrick StoleeAug 25, 2021
  31. Patrick SteinhardtSep 1, 2021
  32. 7/7 fetch: avoid second connectivity check if we already have all objectsPatrick Steinhardt, Aug 24, 2021
  33. 6/7 fetch: merge fetching and consuming refsPatrick Steinhardt, Aug 24, 2021
  34. Derrick StoleeAug 25, 2021
  35. Patrick SteinhardtSep 1, 2021
  36. Junio C HamanoAug 24, 2021
  37. Patrick SteinhardtAug 25, 2021
  38. Derrick StoleeAug 25, 2021
  39. 0/7 Speed up mirror-fetches with many refsPatrick Steinhardt, Sep 1, 2021
  40. 1/7 fetch: speed up lookup of want refs via commit-graphPatrick Steinhardt, Sep 1, 2021
  41. 2/7 fetch: avoid unpacking headers in object existence checkPatrick Steinhardt, Sep 1, 2021
  42. 3/7 connected: refactor iterator to return next object ID directlyPatrick Steinhardt, Sep 1, 2021
  43. 4/7 fetch-pack: optimize loading of refs via commit graphPatrick Steinhardt, Sep 1, 2021
  44. 5/7 fetch: refactor fetch refs to be more extendablePatrick Steinhardt, Sep 1, 2021
  45. 6/7 fetch: merge fetching and consuming refsPatrick Steinhardt, Sep 1, 2021
  46. 7/7 fetch: avoid second connectivity check if we already have all objectsPatrick Steinhardt, Sep 1, 2021
  47. Junio C HamanoSep 1, 2021
  48. Junio C HamanoSep 8, 2021

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.